Catastrophe Vulnerability and Risk Mapping in the Iron Quadrangle, Brazil – Preliminary Results in the Rio das Velhas Watershed Raphael de Vicq1, Teresa Albuquerque2, Rita Fonseca3, Mariangela Garcia Praça Leite4

1Earth Science Institute (ESI), University of Evora, Portugal and Federal University of Ouro Preto, Brazil 2Earth Science Institute (ESI), University of Evora and Polytechnique Institute of Castelo Branco| CERNAS|QRural, Av. Pedro Álvares Cabral, nº 12, 6000-084 Castelo Branco, Portugal, ORCID-0000-0002-8782-6133, [email protected] 3Earth Science Institute (ESI), Department of Geosciences, School of Science and Technology, AmbiTerra Laboratory, University of Évora, Portugal 4Department of Geology, School of Mines, Federal University of Ouro Preto, Brazil

Abstract The geochemistry of fluvial deposits is one of the most used proxies for contamination assessment in mined areas. A river sediment survey was carried out in Iron Quadrangle (IQ), Brazil, between 2013 and 2015. Subsequently, in the last four years, two dam failures happened, causing great environmental damage. Geostatistical modelling was used to model Potentially Toxic Elements’ spatial patterns and the definition of hot/ cold clusters for Arsenic contamination risk concerning catastrophic scenarios, such as dam failure. and so used as a tool for vulnerability and risk assessment. The preliminary results for Arsenic spatial distribution are introduced and discussed. Keywords: Iron Quadrangle, Potentially Toxic Elements, Geostatistical Modelling

Introduction (12.5% of total exports and 36.6% of the trade Fluvial sediments come from a myriad of balance – AMN 2019) and Minas Gerais sources, diffuse and punctual, whose relative State is responsible for the other 40% of the contribution varies over time and space. commercialized mineral production (AMN Their geochemical composition is a response 2019). The Iron Quadrangle (IQ), southeast to the availability of materials, whether due of Minas Gerais State, (fig. 1) is one of the to natural causes or human activities, such as most important mining producing provinces mining. Mining industries produce enormous in Brazil and one of the World’s most volumes of waste material, mainly tailings, important mineral regions, covering an area 2 which can be a relevant source of trace of approximately 7,000 km . Mining activities element contamination for the environment, are focus on iron and gold exploitations, especially in the river basins (Krıbek et al. whose wastes are stored in large tailings 2014). Worldwide, billions of tonnes of dams. The IQ yields the headwaters of two tailings are contained in impoundments major Brazilian River basins: Doce and São behind huge dams (Owen et al. 2020). Several Francisco. The last one, fed by Velhas (largest characteristics of these structures make them drainage basin in IQ) and Paraopeba rivers. more susceptible to failures.The enormous Unfortunately, during the last six years, volume and characteristics of the material IQ experience two of the most dramatic released in the environment with the collapse accidents involving the dam’s tailings failure. of the dam, especially in watersheds and river Fundão dam (Bento Rodrigues/Mariana), 7 3 basins, affect the quality of sediments and on 5 November 2015, released 4.3×10 m water (Kossoff et al. 2014). of tailings in Doce River basin (Carmo Mining is still among the most productive et al. 2017) and Córrego do Feijão dam activities that support the Brazilian economy (Brumadinho), on 25 January 2019, released

Pope, J.; Wolkersdorfer, Ch.; Weber, A.; Sartz, L.; Wolkersdorfer, K. (Editors) 13 IMWA 2020 “Mine Water Solutions”

1.2×107 m3 of tailings in Paraopeba River river basins is mining activity (Deschamps basin (Thompson et al. 2020). & Matschullat, 2011). Several studies have A preliminary multivariate study was identified anomalous concentrations of performed through Principal Component arsenic in IQ waters and sediments (Costa Analysis (PCA) (Shaw 2003) to understand et al. 2015). In the case of sediments, there the attributes of preferential association and are reports of levels above 4700 µg.g-1 (Borba the reduction of the space of analyses. The et al. 2000), much higher than the average analysis begins with p random attributes concentration in the Earth’s crust (between X1, X2, …, Xp, where no assumption 1.0 – 4.8 µg.g-1, according to Taylor & of multivariate normality is required. McLennan, 1995 and Rudnick & Gao, 2003). Considering the ACP outputs, Cr, Zn, Cd, Arsenic is defined as a ‘Regionalized Ni, Cu and Pb were kept for further spatial Variable (Matheron 1971) and consequently analysis as they can act as accurate indicators additive by construction, since the mean for the pollution characterization. value within given observed support is equal In the herein survey the results for the to the arithmetic average of sample values, As spatial distribution are introduced and regardless of the statistical distribution of the discussed as As plays a key role in soil and values. Geostatistical modelling (Goovaerts sediments pollution (Gonzalez-Fernandez 1997, Journel and Huijbregts 1978) was et al. 2018). Indeed, among the elements used, throughout conventional variography present in iron and gold tailings, Arsenic followed by Sequential Gaussian Simulation appears as one of the most dangerous to algorithm (SGS) and local G clustering, the environment, including human health to model Arsenic concentration’s spatial (Fewtrell et al. 2005, Zhang et al., 2019). patterns and the definition of hot/cold Although the interaction between arsenic and spots for contamination risk. The Standard the various environmental compartments Deviation map, obtained from the performed (sediment, soil, water) still needs to be better one hundred simulations (SGS), allowed understood, there is no doubt that one of the the visualization of the correspondent main sources of arsenic contamination in spatial uncertainty and, therefore, acting

Figure 1 Study area location.

14 Pope, J.; Wolkersdorfer, Ch.; Weber, A.; Sartz, L.; Wolkersdorfer, K. (Editors) IMWA 2020 “Mine Water Solutions” as a measurement of the obtained clusters with simple kriging (SK) using the normal robustness and providing a faster and score data and a zero mean (Goovaerts more intuitive way to verify whether the 1997). Once all normal scores had been problematic zones detected previously are simulated, they were back-transformed to true of concern, focusing on the visualization original grade values (Albuquerque et al., and delineation of potential zones for future 2017). For the computation, the Space- monitoring and remediation. Stat Software V. 4.0.18, Biomedwere, was used. The outcome of a simulation is a Methods twisted version of an estimation process, Five hundred and forty-one (541) stream which reproduces the statistics of the sediments were sampled throughout the known data, making a realistic look of the entire IQ (7,000 km2), providing a sampling exemplar, but providing a low prediction density of one sample per 13 km2. The behaviour. If multiple sequences of concentrations of the key PTEs: As; Cd; Co; simulation is designed, it is possible to Cr; Cu; Ni; Pb and Zn and associated metals obtain more reliable probabilistic maps; (Fe, Mn) were obtained through aqua regia 3. Finally, Local G clustering allowed digestion followed by ICP-AES (Spectro measurement of the degree of association Ciros CCD) analysis. that results from the concentration of A three-stepgeostatistical modelling weighted points (or region represented by methodology was used for the construction a weighted point) and all other weighted of the Arsenic spatial distribution maps and points included within a radius of distance the definition of spatial hot-spots, as follows: from the original (Getis and Ord 1992). 1. Selected attributes went through structural analysis and experimental When predicting the risk of contamination variograms were computed. The(e.g. months ahead), it is mandatory to variogram is a vector function used to stress the relevance of the chances for the compute the spatial variation structure of future estimated values exceeding maximum regionalized variables (Matheron 1971; admissible values. The delineation of zones of Journel and Huijbregts 1978); high and low risk requires the interpolation 2. Sequential Gaussian Simulation (SGS) of risk values to the nodes of a regular grid was used as a stochastic simulation making possible proper risk assessments, algorithm. SGS starts by defining and a prediction model working as guidance the univariate distribution of values, to a more sustainable environmental performing a normal score transform of management. The three watersheds (Doce, the original values to a standard normal Velhas, and Paraopeba) were processed distribution. Normal scores at grid node separately (fig. 2) guarantying that only the locations were simulated sequentially samples inside each geographic envelop were

Doce Watershed Velhas Watershed

Paraopeba Watershed

Figure 2 Arsenic Spatial Distribution – SGS mean image.

Pope, J.; Wolkersdorfer, Ch.; Weber, A.; Sartz, L.; Wolkersdorfer, K. (Editors) 15 IMWA 2020 “Mine Water Solutions”

Figure 3 Arsenic Spatial Distribution – SGS mean image and spatial uncertainty (standard deviation). used for the correspondent spatial modelling. is possible to identify high rings (hot spots Finally, all the representations were projected for As pollution) and low rings (cold spots for together for visualization purposes (fig. 3 and As pollution) pointing out to the Velhas and fig. 4). Observing the three surveyed units, Doce watersheds as the ones in need of close Velhas, Doce, and Paraopeba watersheds, monitoring. it is possible to acknowledge, for As spatial Still, a source of countless debates, distribution, the low risk in the Paraopeba the anthropogenic contribution to high and the high risk observed mostly in the concentrations of arsenic in fluvial sediments central area of the Velhas basin. is not well defined. Even so, establish the It is also worth notice the high spatial natural abundance of an element is essential uncertainty associated with the Velhas’ basin to support any type of environmental data, mainly due to the presence of several analysis. Concerning sediments, global or severe outliers within the studied area (fig. 3). regional standards can be used. Arsenic Seventeen percent of As concentration values average concentrations calculated for QF are higher than the mean (32.33 µg.g-1) and (18.17 µg.g1), Velhas river basin (32.33 µg.g- 1) 6,7% higher or equal to 100 µg.g1. and Doce river basin (14.23 µg.g-1) are Finally, when observing the G-clusters it considerably higher than the highest value

Figure 4 Arsenic Spatial G-Clusters.

16 Pope, J.; Wolkersdorfer, Ch.; Weber, A.; Sartz, L.; Wolkersdorfer, K. (Editors) IMWA 2020 “Mine Water Solutions” set for the Earth’s crust (4.8 µg.g-1 – Rudnick Conclusions & Gao 2003). Only in the Paraopeba river The stochastic modelling of the arsenic watershed As mean concentration is lower concentrations in the fluvial sediments -1 (4.01 µg.g ). Besides that, Costa et al. in the IQ allowed to identify the Velhas (2015) defined the natural background of River watershed as the most vulnerable -1 As concentration in IQ as 6.1 µg.g and area to catastrophic risk given the high -1 values >12.8 µg.g as anomalies. Regardless concentrations of Arsenic in this basin, with of the limits considered, Velhas and Doce an average of 32.33 µg.g-1, about 16 times river basins can be considered areas with above the upper crustal average, showing that anomalous values of As concentrations. there is a significant amount of this element Several studies pointed out that in scattered throughout the watershed. IQ, arsenic concentrations in water and In this basin there are 18 gold mines, six in sediments are related to gold deposits, operation and 12 paralyzed, (Pinto and Silva, present in minerals such as arsenopyrite 2014), which extracted approximately 692 tons and arseniferous pyrites (Borba et al. 2000). of gold over the last 40 years (Goldfarb and Gold can be found in veins of quartz and Groves 2015, Codemge 2017), considering carbonates from the Nova Lima Group (base the Au/As ratio presented by Costa (2007), of the Rio das Velhas Supergroup) and base it is estimated that 1,868,400 tons of As, are of the Minas Supergroup (Deschamps and potentially scattered in the basin or stored Matschullat, 2011). Moreover, gold mine in tailings piles, including abandoned mines waste (tailings) containing arsenic have been and waste pile, which represents a major released to the environment or store in tailing concern in catastrophic scenarios, which are dams for the last three centuries (Matschullat identified in this research. et al. 2000, Borba et al. 2000). Although iron Although recent dam breaks have -1 ores usually have As concentrations <1 µg.g , occurred in the Doce river basin and the sorbed, or coprecipitated in iron (oxy)(hydr) Paraopeba River Basin, the average As oxides, their tailings can be enriched in As. concentrations in these areas, when compared Indeed, it is possible to identify significant- with Velhas river basin, are much lower, high rings in the northern and southeast 4.01 µg.g-1 (8 times below) in the Paraopeba areas of the Velhas River watershed, all river and 14.23 µg.g- 1 (2.3 times below) in the famous locations for primary gold deposits Doce River, which points to its higher risk in and gold mines. Considering the Au/As the event of the failure of one or more tailings ratio presented by Costa (2007) for gold dams, as in the other two basins. ore exploitations in IQ, almost 2,000,000 Thus, it was observed that in the Rio das tons of As are potentially scattered in the Velhas Basin there is a greater catastrophe basin or stored in waste piles and tailings vulnerability, which confirms the need to dams, including abandoned mines, which apply geostatistical modeling as a tool to represents a major concern in catastrophic identify these areas and spatial patterns of scenarios identified in this research. Besides PTEs, supporting government agencies in the that, in Velhas river basin there are, actually, prevention of environmental damage and the 16 iron tailing dams, five of then with a high execution of more accurate monitoring. risk of failure. It should be noted that in the high ring identified in the north of figure 3, Acknowledgments currently, 450,000 inhabitants live, in addition The authors thank all co-organisers for to hosting the main source of water supply for hosting the IMWA2020 Conference. Amy the city of Belo Horizonte, the state capital, Kokoska, Hetta Pieterse as well as Glenn with 2,500,000 inhabitants. Three of the five MacLeod provided critical comments on dams with a risk of failure are localized in earlier versions of this text. The research or upstream this area, and part of the people was supported by (1) CERNAS-IPCB living downstream from these structures have [UIDB/00681/2020 funding by Foundation already been removed. for Science and Technology (FCT), by (2) ICT, under contract with FCT (the Portuguese

Pope, J.; Wolkersdorfer, Ch.; Weber, A.; Sartz, L.; Wolkersdorfer, K. (Editors) 17 IMWA 2020 “Mine Water Solutions”

Science and Technology Foundation) estimation of the global burden of disease due to through the Projects UIDB/04683/2020 e skin lesions caused by arsenic in drinking water. UIDP/04683/2020 and by (3) FAPEMIG Journal of Water and Health, 3(2), 101–107. (Research supporting foundation of Minas Getis A, Ord JK (1992) The analysis of spatial Gerais State). association by use of distance statistics. Geogr. Anal. 24(3): 189–206 References Goldfarb R.J. & Groves D.I. (2015). Orogenic gold: Albuquerque MTD., Gerassis S, Sierra C, Taboada common or evolving fluid and metal sources J, Martín JE., Antunes IMHR, Gallego JR. (2017) through time. Lithos, 233:2-26. doi: 10.1016/j. Developing a new Bayesian Risk Index for risk lithos.2015.07.011 evaluation of soil contamination. Sci Total Gonzalez-Fernandez B, Rodríguez-Valdes E, Environ 603–604:, 1167–177, doi: 10.1016/j. Boente C., Menendez-Casares E, Fernandez- scitotenv.2017.06.068 Braña A, Gallego JR. (2018) Long-term ANM - Agência Nacional de Mineração. ongoing impact of arsenic contamination (2019) Anuário Mineral Brasileiro Principais on the environmental compartments of a Substâncias Metálicas 2019, former mining metallurgy area. Sci. Total http://www.anm.gov.br/dnpm/publicacoes/ Environ. 610-611: 820-830. doi: 10.1016/j. serie-estatisticas-e-economia-mineral/ scitotenv.2017.08.135 anuario-mineral/anuario-mineral-brasileiro/ Goovaerts P (1997) Geostatistics for Natural amb_2019_ano_base_2018 Resources Evaluation. University Press, New Borba RP, Figueiredo BR, Rawlins B, Matschullat York: Oxford J (2000) Arsenic in Water and Sediment in the Journel A, Huijbregts CJ (1978) Mining Quadrilátero Ferrífero, State of Minas Gerais, Geostatistics. Academic Press, San Diego Brazil. Applied Geochemistry, v. 15, n. 2: 181-190 Kossoff D, Dubbin WE, Alfredsson M, Edwards Carmo, FF, Kamino, LHY, Junior, RT, de Campos, SJ, Macklin MG, Hudson-Edwards, KA (2014) IC, do Carmo, FF, Silvino, G, de Castro, KJSX. Mine tailings dams: characteristics, failure, Mauro, ML, Rodrigues, NUA, Miranda, environmental impacts, and remediation Appl. MPS, Pinto CEF (2017) Fundão tailings dam Geochem., 51:229-245 failures: the environment tragedy of the largest Krıbek B, Davies TC, De Vivo B (2014) Mining technological disaster of Brazilian mining vs. environment in Africa. J Geochem Explor in global context Perspect. Ecol. Conserv., 144:387–580 15:145-151 Matheron G (1971) The Theory of Regionalized CODEMGE - Companhia de Desenvolvimento Variables and Its Applications. Edição 5 de Minas Gerais. (2018) Recursos minerais de Les Cahiers du Centre de morphologie de Minas Gerais. http://recursomineralmg. mathématique de Fontainebleau, Centre de codemge.com.br/substancias-minerais/ouro/# morphologie mathématique de Fontainebleau, Costa AT (2007) Registro histórico de ISSN 1157-1195 contaminacão por metais pesados, associadas Mattschullat J, Borba RP, Deschamps E, à` exploração aurífera no Alto e Médio curso da Figueiredo BF, Gabrio T, Schwenk M (2000) Bacia do Ribeirão do Carmo, QF: um estudo de Human and environmental contamination sedimentos de planícies de inundacão e terraços in the Quadrilátero Ferrífero, Brazil. Applied aluviais. Doctoral thesis, UFOP, Ouro Preto. Geochemistry, v. 15, p. 181-190. 10.1016/S0883- Costa RVF, Leite MGP, Mendonça FPC, Nalini 2927(99)00039-6 JHA (2015) Geochemical mapping of arsenic Owen JR, Kemp D, Lèbre É, Svobodova, Pérez in surface waters and stream sediments of the Murillo, KG (2020) Catastrophic tailings Quadrilátero Ferrífero, Brazil Geociências, doi: dam failures and disaster risk disclosure. 10.1590/0370-44672015680077 International Journal of Disaster Risk Reduction Deschamps E, Matschullat J (eds) (2011) Arsenic: 42:101361 natural and anthropogenic. Arsenic in the Pinto CP, Silva MA (2014). Mapa Geológico do Environment 4, p 209. CRC Press, Boca Raton Estado de Minas Gerais, Escala 1:1.000.000. Fewtrell L., Fuge, R., & Kay, D. (2005). An Companhia de Desenvolvimento Econômico de

18 Pope, J.; Wolkersdorfer, Ch.; Weber, A.; Sartz, L.; Wolkersdorfer, K. (Editors) IMWA 2020 “Mine Water Solutions”

Minas Gerais, CODEMIG e Serviço Geológico Thompson F, Oliveira BC, Cordeiro MC, Masi BP, do Brasil, CPRM. Rangel TP, Paz P, Freitas T, Lopes G, Silva BS, Robb L (2004) Introduction to Ore-forming Cabral AS, Soares M, Lacerda D, Vergilio CS, Processes. Wiley-Blackwell, Oxford. Ferreira ML, Lima C, Thompson C, Rezende Shaw PJA (2003) Multivariate statistics for the CE (2020). Severe impacts of the Brumadinho Environmental Sciences, Hodder-Arnold. ISBN dam failure (Minas Gerais, Brazil) on the water 0-340-80763-6 quality of the Paraopeba River. Science of The Rudnick RL, Gao S (2003) Composition of the Total Environment, 705, 1-24. doi.org/10.1016/j. continental crust. In The Crust, Treatise on scitotenv.2019.135914. Geochemistry, Vol. 3, (ed. Rudnick, R.L.). Zhang Y, Xu B, Guo Z, Han J, Li H, Jim L, Chen F, Elsevier, 1-64. Xiong Y (2019) Human health risk assessment Taylor SR, McLennan SM (1995) The geochemical of groundwater arsenic contamination in evolution of the continental crust. Rev Geophys; Jinghui irrigation district, China. J. Environ. 33: 241-265. Manag., 237:163-169

Pope, J.; Wolkersdorfer, Ch.; Weber, A.; Sartz, L.; Wolkersdorfer, K. (Editors) 19